1 //===-- LoopUtils.cpp - Loop Utility functions -------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines common loop utility functions.
12 //===----------------------------------------------------------------------===//
14 #include "llvm/Transforms/Utils/LoopUtils.h"
15 #include "llvm/ADT/ScopeExit.h"
16 #include "llvm/Analysis/AliasAnalysis.h"
17 #include "llvm/Analysis/BasicAliasAnalysis.h"
18 #include "llvm/Analysis/GlobalsModRef.h"
19 #include "llvm/Analysis/LoopInfo.h"
20 #include "llvm/Analysis/LoopPass.h"
21 #include "llvm/Analysis/ScalarEvolution.h"
22 #include "llvm/Analysis/ScalarEvolutionAliasAnalysis.h"
23 #include "llvm/Analysis/ScalarEvolutionExpander.h"
24 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
25 #include "llvm/Analysis/TargetTransformInfo.h"
26 #include "llvm/IR/Dominators.h"
27 #include "llvm/IR/Instructions.h"
28 #include "llvm/IR/Module.h"
29 #include "llvm/IR/PatternMatch.h"
30 #include "llvm/IR/ValueHandle.h"
31 #include "llvm/Pass.h"
32 #include "llvm/Support/Debug.h"
33 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
36 using namespace llvm::PatternMatch;
38 #define DEBUG_TYPE "loop-utils"
40 bool RecurrenceDescriptor::areAllUsesIn(Instruction *I,
41 SmallPtrSetImpl<Instruction *> &Set) {
42 for (User::op_iterator Use = I->op_begin(), E = I->op_end(); Use != E; ++Use)
43 if (!Set.count(dyn_cast<Instruction>(*Use)))
48 bool RecurrenceDescriptor::isIntegerRecurrenceKind(RecurrenceKind Kind) {
57 case RK_IntegerMinMax:
63 bool RecurrenceDescriptor::isFloatingPointRecurrenceKind(RecurrenceKind Kind) {
64 return (Kind != RK_NoRecurrence) && !isIntegerRecurrenceKind(Kind);
67 bool RecurrenceDescriptor::isArithmeticRecurrenceKind(RecurrenceKind Kind) {
81 RecurrenceDescriptor::lookThroughAnd(PHINode *Phi, Type *&RT,
82 SmallPtrSetImpl<Instruction *> &Visited,
83 SmallPtrSetImpl<Instruction *> &CI) {
84 if (!Phi->hasOneUse())
87 const APInt *M = nullptr;
88 Instruction *I, *J = cast<Instruction>(Phi->use_begin()->getUser());
90 // Matches either I & 2^x-1 or 2^x-1 & I. If we find a match, we update RT
91 // with a new integer type of the corresponding bit width.
92 if (match(J, m_c_And(m_Instruction(I), m_APInt(M)))) {
93 int32_t Bits = (*M + 1).exactLogBase2();
95 RT = IntegerType::get(Phi->getContext(), Bits);
104 bool RecurrenceDescriptor::getSourceExtensionKind(
105 Instruction *Start, Instruction *Exit, Type *RT, bool &IsSigned,
106 SmallPtrSetImpl<Instruction *> &Visited,
107 SmallPtrSetImpl<Instruction *> &CI) {
109 SmallVector<Instruction *, 8> Worklist;
110 bool FoundOneOperand = false;
111 unsigned DstSize = RT->getPrimitiveSizeInBits();
112 Worklist.push_back(Exit);
114 // Traverse the instructions in the reduction expression, beginning with the
116 while (!Worklist.empty()) {
117 Instruction *I = Worklist.pop_back_val();
118 for (Use &U : I->operands()) {
120 // Terminate the traversal if the operand is not an instruction, or we
121 // reach the starting value.
122 Instruction *J = dyn_cast<Instruction>(U.get());
123 if (!J || J == Start)
126 // Otherwise, investigate the operation if it is also in the expression.
127 if (Visited.count(J)) {
128 Worklist.push_back(J);
132 // If the operand is not in Visited, it is not a reduction operation, but
133 // it does feed into one. Make sure it is either a single-use sign- or
134 // zero-extend instruction.
135 CastInst *Cast = dyn_cast<CastInst>(J);
136 bool IsSExtInst = isa<SExtInst>(J);
137 if (!Cast || !Cast->hasOneUse() || !(isa<ZExtInst>(J) || IsSExtInst))
140 // Ensure the source type of the extend is no larger than the reduction
141 // type. It is not necessary for the types to be identical.
142 unsigned SrcSize = Cast->getSrcTy()->getPrimitiveSizeInBits();
143 if (SrcSize > DstSize)
146 // Furthermore, ensure that all such extends are of the same kind.
147 if (FoundOneOperand) {
148 if (IsSigned != IsSExtInst)
151 FoundOneOperand = true;
152 IsSigned = IsSExtInst;
155 // Lastly, if the source type of the extend matches the reduction type,
156 // add the extend to CI so that we can avoid accounting for it in the
158 if (SrcSize == DstSize)
165 bool RecurrenceDescriptor::AddReductionVar(PHINode *Phi, RecurrenceKind Kind,
166 Loop *TheLoop, bool HasFunNoNaNAttr,
167 RecurrenceDescriptor &RedDes) {
168 if (Phi->getNumIncomingValues() != 2)
171 // Reduction variables are only found in the loop header block.
172 if (Phi->getParent() != TheLoop->getHeader())
175 // Obtain the reduction start value from the value that comes from the loop
177 Value *RdxStart = Phi->getIncomingValueForBlock(TheLoop->getLoopPreheader());
179 // ExitInstruction is the single value which is used outside the loop.
180 // We only allow for a single reduction value to be used outside the loop.
181 // This includes users of the reduction, variables (which form a cycle
182 // which ends in the phi node).
183 Instruction *ExitInstruction = nullptr;
184 // Indicates that we found a reduction operation in our scan.
185 bool FoundReduxOp = false;
187 // We start with the PHI node and scan for all of the users of this
188 // instruction. All users must be instructions that can be used as reduction
189 // variables (such as ADD). We must have a single out-of-block user. The cycle
190 // must include the original PHI.
191 bool FoundStartPHI = false;
193 // To recognize min/max patterns formed by a icmp select sequence, we store
194 // the number of instruction we saw from the recognized min/max pattern,
195 // to make sure we only see exactly the two instructions.
196 unsigned NumCmpSelectPatternInst = 0;
197 InstDesc ReduxDesc(false, nullptr);
199 // Data used for determining if the recurrence has been type-promoted.
200 Type *RecurrenceType = Phi->getType();
201 SmallPtrSet<Instruction *, 4> CastInsts;
202 Instruction *Start = Phi;
203 bool IsSigned = false;
205 SmallPtrSet<Instruction *, 8> VisitedInsts;
206 SmallVector<Instruction *, 8> Worklist;
208 // Return early if the recurrence kind does not match the type of Phi. If the
209 // recurrence kind is arithmetic, we attempt to look through AND operations
210 // resulting from the type promotion performed by InstCombine. Vector
211 // operations are not limited to the legal integer widths, so we may be able
212 // to evaluate the reduction in the narrower width.
213 if (RecurrenceType->isFloatingPointTy()) {
214 if (!isFloatingPointRecurrenceKind(Kind))
217 if (!isIntegerRecurrenceKind(Kind))
219 if (isArithmeticRecurrenceKind(Kind))
220 Start = lookThroughAnd(Phi, RecurrenceType, VisitedInsts, CastInsts);
223 Worklist.push_back(Start);
224 VisitedInsts.insert(Start);
226 // A value in the reduction can be used:
227 // - By the reduction:
228 // - Reduction operation:
229 // - One use of reduction value (safe).
230 // - Multiple use of reduction value (not safe).
232 // - All uses of the PHI must be the reduction (safe).
233 // - Otherwise, not safe.
234 // - By instructions outside of the loop (safe).
235 // * One value may have several outside users, but all outside
236 // uses must be of the same value.
237 // - By an instruction that is not part of the reduction (not safe).
239 // * An instruction type other than PHI or the reduction operation.
240 // * A PHI in the header other than the initial PHI.
241 while (!Worklist.empty()) {
242 Instruction *Cur = Worklist.back();
246 // If the instruction has no users then this is a broken chain and can't be
247 // a reduction variable.
248 if (Cur->use_empty())
251 bool IsAPhi = isa<PHINode>(Cur);
253 // A header PHI use other than the original PHI.
254 if (Cur != Phi && IsAPhi && Cur->getParent() == Phi->getParent())
257 // Reductions of instructions such as Div, and Sub is only possible if the
258 // LHS is the reduction variable.
259 if (!Cur->isCommutative() && !IsAPhi && !isa<SelectInst>(Cur) &&
260 !isa<ICmpInst>(Cur) && !isa<FCmpInst>(Cur) &&
261 !VisitedInsts.count(dyn_cast<Instruction>(Cur->getOperand(0))))
264 // Any reduction instruction must be of one of the allowed kinds. We ignore
265 // the starting value (the Phi or an AND instruction if the Phi has been
268 ReduxDesc = isRecurrenceInstr(Cur, Kind, ReduxDesc, HasFunNoNaNAttr);
269 if (!ReduxDesc.isRecurrence())
273 // A reduction operation must only have one use of the reduction value.
274 if (!IsAPhi && Kind != RK_IntegerMinMax && Kind != RK_FloatMinMax &&
275 hasMultipleUsesOf(Cur, VisitedInsts))
278 // All inputs to a PHI node must be a reduction value.
279 if (IsAPhi && Cur != Phi && !areAllUsesIn(Cur, VisitedInsts))
282 if (Kind == RK_IntegerMinMax &&
283 (isa<ICmpInst>(Cur) || isa<SelectInst>(Cur)))
284 ++NumCmpSelectPatternInst;
285 if (Kind == RK_FloatMinMax && (isa<FCmpInst>(Cur) || isa<SelectInst>(Cur)))
286 ++NumCmpSelectPatternInst;
288 // Check whether we found a reduction operator.
289 FoundReduxOp |= !IsAPhi && Cur != Start;
291 // Process users of current instruction. Push non-PHI nodes after PHI nodes
292 // onto the stack. This way we are going to have seen all inputs to PHI
293 // nodes once we get to them.
294 SmallVector<Instruction *, 8> NonPHIs;
295 SmallVector<Instruction *, 8> PHIs;
296 for (User *U : Cur->users()) {
297 Instruction *UI = cast<Instruction>(U);
299 // Check if we found the exit user.
300 BasicBlock *Parent = UI->getParent();
301 if (!TheLoop->contains(Parent)) {
302 // If we already know this instruction is used externally, move on to
304 if (ExitInstruction == Cur)
307 // Exit if you find multiple values used outside or if the header phi
308 // node is being used. In this case the user uses the value of the
309 // previous iteration, in which case we would loose "VF-1" iterations of
310 // the reduction operation if we vectorize.
311 if (ExitInstruction != nullptr || Cur == Phi)
314 // The instruction used by an outside user must be the last instruction
315 // before we feed back to the reduction phi. Otherwise, we loose VF-1
316 // operations on the value.
317 if (!is_contained(Phi->operands(), Cur))
320 ExitInstruction = Cur;
324 // Process instructions only once (termination). Each reduction cycle
325 // value must only be used once, except by phi nodes and min/max
326 // reductions which are represented as a cmp followed by a select.
327 InstDesc IgnoredVal(false, nullptr);
328 if (VisitedInsts.insert(UI).second) {
329 if (isa<PHINode>(UI))
332 NonPHIs.push_back(UI);
333 } else if (!isa<PHINode>(UI) &&
334 ((!isa<FCmpInst>(UI) && !isa<ICmpInst>(UI) &&
335 !isa<SelectInst>(UI)) ||
336 !isMinMaxSelectCmpPattern(UI, IgnoredVal).isRecurrence()))
339 // Remember that we completed the cycle.
341 FoundStartPHI = true;
343 Worklist.append(PHIs.begin(), PHIs.end());
344 Worklist.append(NonPHIs.begin(), NonPHIs.end());
347 // This means we have seen one but not the other instruction of the
348 // pattern or more than just a select and cmp.
349 if ((Kind == RK_IntegerMinMax || Kind == RK_FloatMinMax) &&
350 NumCmpSelectPatternInst != 2)
353 if (!FoundStartPHI || !FoundReduxOp || !ExitInstruction)
356 // If we think Phi may have been type-promoted, we also need to ensure that
357 // all source operands of the reduction are either SExtInsts or ZEstInsts. If
358 // so, we will be able to evaluate the reduction in the narrower bit width.
360 if (!getSourceExtensionKind(Start, ExitInstruction, RecurrenceType,
361 IsSigned, VisitedInsts, CastInsts))
364 // We found a reduction var if we have reached the original phi node and we
365 // only have a single instruction with out-of-loop users.
367 // The ExitInstruction(Instruction which is allowed to have out-of-loop users)
368 // is saved as part of the RecurrenceDescriptor.
370 // Save the description of this reduction variable.
371 RecurrenceDescriptor RD(
372 RdxStart, ExitInstruction, Kind, ReduxDesc.getMinMaxKind(),
373 ReduxDesc.getUnsafeAlgebraInst(), RecurrenceType, IsSigned, CastInsts);
379 /// Returns true if the instruction is a Select(ICmp(X, Y), X, Y) instruction
380 /// pattern corresponding to a min(X, Y) or max(X, Y).
381 RecurrenceDescriptor::InstDesc
382 RecurrenceDescriptor::isMinMaxSelectCmpPattern(Instruction *I, InstDesc &Prev) {
384 assert((isa<ICmpInst>(I) || isa<FCmpInst>(I) || isa<SelectInst>(I)) &&
385 "Expect a select instruction");
386 Instruction *Cmp = nullptr;
387 SelectInst *Select = nullptr;
389 // We must handle the select(cmp()) as a single instruction. Advance to the
391 if ((Cmp = dyn_cast<ICmpInst>(I)) || (Cmp = dyn_cast<FCmpInst>(I))) {
392 if (!Cmp->hasOneUse() || !(Select = dyn_cast<SelectInst>(*I->user_begin())))
393 return InstDesc(false, I);
394 return InstDesc(Select, Prev.getMinMaxKind());
397 // Only handle single use cases for now.
398 if (!(Select = dyn_cast<SelectInst>(I)))
399 return InstDesc(false, I);
400 if (!(Cmp = dyn_cast<ICmpInst>(I->getOperand(0))) &&
401 !(Cmp = dyn_cast<FCmpInst>(I->getOperand(0))))
402 return InstDesc(false, I);
403 if (!Cmp->hasOneUse())
404 return InstDesc(false, I);
409 // Look for a min/max pattern.
410 if (m_UMin(m_Value(CmpLeft), m_Value(CmpRight)).match(Select))
411 return InstDesc(Select, MRK_UIntMin);
412 else if (m_UMax(m_Value(CmpLeft), m_Value(CmpRight)).match(Select))
413 return InstDesc(Select, MRK_UIntMax);
414 else if (m_SMax(m_Value(CmpLeft), m_Value(CmpRight)).match(Select))
415 return InstDesc(Select, MRK_SIntMax);
416 else if (m_SMin(m_Value(CmpLeft), m_Value(CmpRight)).match(Select))
417 return InstDesc(Select, MRK_SIntMin);
418 else if (m_OrdFMin(m_Value(CmpLeft), m_Value(CmpRight)).match(Select))
419 return InstDesc(Select, MRK_FloatMin);
420 else if (m_OrdFMax(m_Value(CmpLeft), m_Value(CmpRight)).match(Select))
421 return InstDesc(Select, MRK_FloatMax);
422 else if (m_UnordFMin(m_Value(CmpLeft), m_Value(CmpRight)).match(Select))
423 return InstDesc(Select, MRK_FloatMin);
424 else if (m_UnordFMax(m_Value(CmpLeft), m_Value(CmpRight)).match(Select))
425 return InstDesc(Select, MRK_FloatMax);
427 return InstDesc(false, I);
430 RecurrenceDescriptor::InstDesc
431 RecurrenceDescriptor::isRecurrenceInstr(Instruction *I, RecurrenceKind Kind,
432 InstDesc &Prev, bool HasFunNoNaNAttr) {
433 bool FP = I->getType()->isFloatingPointTy();
434 Instruction *UAI = Prev.getUnsafeAlgebraInst();
435 if (!UAI && FP && !I->hasUnsafeAlgebra())
436 UAI = I; // Found an unsafe (unvectorizable) algebra instruction.
438 switch (I->getOpcode()) {
440 return InstDesc(false, I);
441 case Instruction::PHI:
442 return InstDesc(I, Prev.getMinMaxKind(), Prev.getUnsafeAlgebraInst());
443 case Instruction::Sub:
444 case Instruction::Add:
445 return InstDesc(Kind == RK_IntegerAdd, I);
446 case Instruction::Mul:
447 return InstDesc(Kind == RK_IntegerMult, I);
448 case Instruction::And:
449 return InstDesc(Kind == RK_IntegerAnd, I);
450 case Instruction::Or:
451 return InstDesc(Kind == RK_IntegerOr, I);
452 case Instruction::Xor:
453 return InstDesc(Kind == RK_IntegerXor, I);
454 case Instruction::FMul:
455 return InstDesc(Kind == RK_FloatMult, I, UAI);
456 case Instruction::FSub:
457 case Instruction::FAdd:
458 return InstDesc(Kind == RK_FloatAdd, I, UAI);
459 case Instruction::FCmp:
460 case Instruction::ICmp:
461 case Instruction::Select:
462 if (Kind != RK_IntegerMinMax &&
463 (!HasFunNoNaNAttr || Kind != RK_FloatMinMax))
464 return InstDesc(false, I);
465 return isMinMaxSelectCmpPattern(I, Prev);
469 bool RecurrenceDescriptor::hasMultipleUsesOf(
470 Instruction *I, SmallPtrSetImpl<Instruction *> &Insts) {
471 unsigned NumUses = 0;
472 for (User::op_iterator Use = I->op_begin(), E = I->op_end(); Use != E;
474 if (Insts.count(dyn_cast<Instruction>(*Use)))
482 bool RecurrenceDescriptor::isReductionPHI(PHINode *Phi, Loop *TheLoop,
483 RecurrenceDescriptor &RedDes) {
485 BasicBlock *Header = TheLoop->getHeader();
486 Function &F = *Header->getParent();
487 bool HasFunNoNaNAttr =
488 F.getFnAttribute("no-nans-fp-math").getValueAsString() == "true";
490 if (AddReductionVar(Phi, RK_IntegerAdd, TheLoop, HasFunNoNaNAttr, RedDes)) {
491 DEBUG(dbgs() << "Found an ADD reduction PHI." << *Phi << "\n");
494 if (AddReductionVar(Phi, RK_IntegerMult, TheLoop, HasFunNoNaNAttr, RedDes)) {
495 DEBUG(dbgs() << "Found a MUL reduction PHI." << *Phi << "\n");
498 if (AddReductionVar(Phi, RK_IntegerOr, TheLoop, HasFunNoNaNAttr, RedDes)) {
499 DEBUG(dbgs() << "Found an OR reduction PHI." << *Phi << "\n");
502 if (AddReductionVar(Phi, RK_IntegerAnd, TheLoop, HasFunNoNaNAttr, RedDes)) {
503 DEBUG(dbgs() << "Found an AND reduction PHI." << *Phi << "\n");
506 if (AddReductionVar(Phi, RK_IntegerXor, TheLoop, HasFunNoNaNAttr, RedDes)) {
507 DEBUG(dbgs() << "Found a XOR reduction PHI." << *Phi << "\n");
510 if (AddReductionVar(Phi, RK_IntegerMinMax, TheLoop, HasFunNoNaNAttr,
512 DEBUG(dbgs() << "Found a MINMAX reduction PHI." << *Phi << "\n");
515 if (AddReductionVar(Phi, RK_FloatMult, TheLoop, HasFunNoNaNAttr, RedDes)) {
516 DEBUG(dbgs() << "Found an FMult reduction PHI." << *Phi << "\n");
519 if (AddReductionVar(Phi, RK_FloatAdd, TheLoop, HasFunNoNaNAttr, RedDes)) {
520 DEBUG(dbgs() << "Found an FAdd reduction PHI." << *Phi << "\n");
523 if (AddReductionVar(Phi, RK_FloatMinMax, TheLoop, HasFunNoNaNAttr, RedDes)) {
524 DEBUG(dbgs() << "Found an float MINMAX reduction PHI." << *Phi << "\n");
527 // Not a reduction of known type.
531 bool RecurrenceDescriptor::isFirstOrderRecurrence(
532 PHINode *Phi, Loop *TheLoop,
533 DenseMap<Instruction *, Instruction *> &SinkAfter, DominatorTree *DT) {
535 // Ensure the phi node is in the loop header and has two incoming values.
536 if (Phi->getParent() != TheLoop->getHeader() ||
537 Phi->getNumIncomingValues() != 2)
540 // Ensure the loop has a preheader and a single latch block. The loop
541 // vectorizer will need the latch to set up the next iteration of the loop.
542 auto *Preheader = TheLoop->getLoopPreheader();
543 auto *Latch = TheLoop->getLoopLatch();
544 if (!Preheader || !Latch)
547 // Ensure the phi node's incoming blocks are the loop preheader and latch.
548 if (Phi->getBasicBlockIndex(Preheader) < 0 ||
549 Phi->getBasicBlockIndex(Latch) < 0)
552 // Get the previous value. The previous value comes from the latch edge while
553 // the initial value comes form the preheader edge.
554 auto *Previous = dyn_cast<Instruction>(Phi->getIncomingValueForBlock(Latch));
555 if (!Previous || !TheLoop->contains(Previous) || isa<PHINode>(Previous) ||
556 SinkAfter.count(Previous)) // Cannot rely on dominance due to motion.
559 // Ensure every user of the phi node is dominated by the previous value.
560 // The dominance requirement ensures the loop vectorizer will not need to
561 // vectorize the initial value prior to the first iteration of the loop.
562 // TODO: Consider extending this sinking to handle other kinds of instructions
563 // and expressions, beyond sinking a single cast past Previous.
564 if (Phi->hasOneUse()) {
565 auto *I = Phi->user_back();
566 if (I->isCast() && (I->getParent() == Phi->getParent()) && I->hasOneUse() &&
567 DT->dominates(Previous, I->user_back())) {
568 SinkAfter[I] = Previous;
573 for (User *U : Phi->users())
574 if (auto *I = dyn_cast<Instruction>(U)) {
575 if (!DT->dominates(Previous, I))
582 /// This function returns the identity element (or neutral element) for
584 Constant *RecurrenceDescriptor::getRecurrenceIdentity(RecurrenceKind K,
590 // Adding, Xoring, Oring zero to a number does not change it.
591 return ConstantInt::get(Tp, 0);
593 // Multiplying a number by 1 does not change it.
594 return ConstantInt::get(Tp, 1);
596 // AND-ing a number with an all-1 value does not change it.
597 return ConstantInt::get(Tp, -1, true);
599 // Multiplying a number by 1 does not change it.
600 return ConstantFP::get(Tp, 1.0L);
602 // Adding zero to a number does not change it.
603 return ConstantFP::get(Tp, 0.0L);
605 llvm_unreachable("Unknown recurrence kind");
609 /// This function translates the recurrence kind to an LLVM binary operator.
610 unsigned RecurrenceDescriptor::getRecurrenceBinOp(RecurrenceKind Kind) {
613 return Instruction::Add;
615 return Instruction::Mul;
617 return Instruction::Or;
619 return Instruction::And;
621 return Instruction::Xor;
623 return Instruction::FMul;
625 return Instruction::FAdd;
626 case RK_IntegerMinMax:
627 return Instruction::ICmp;
629 return Instruction::FCmp;
631 llvm_unreachable("Unknown recurrence operation");
635 Value *RecurrenceDescriptor::createMinMaxOp(IRBuilder<> &Builder,
636 MinMaxRecurrenceKind RK,
637 Value *Left, Value *Right) {
638 CmpInst::Predicate P = CmpInst::ICMP_NE;
641 llvm_unreachable("Unknown min/max recurrence kind");
643 P = CmpInst::ICMP_ULT;
646 P = CmpInst::ICMP_UGT;
649 P = CmpInst::ICMP_SLT;
652 P = CmpInst::ICMP_SGT;
655 P = CmpInst::FCMP_OLT;
658 P = CmpInst::FCMP_OGT;
662 // We only match FP sequences with unsafe algebra, so we can unconditionally
663 // set it on any generated instructions.
664 IRBuilder<>::FastMathFlagGuard FMFG(Builder);
666 FMF.setUnsafeAlgebra();
667 Builder.setFastMathFlags(FMF);
670 if (RK == MRK_FloatMin || RK == MRK_FloatMax)
671 Cmp = Builder.CreateFCmp(P, Left, Right, "rdx.minmax.cmp");
673 Cmp = Builder.CreateICmp(P, Left, Right, "rdx.minmax.cmp");
675 Value *Select = Builder.CreateSelect(Cmp, Left, Right, "rdx.minmax.select");
679 InductionDescriptor::InductionDescriptor(Value *Start, InductionKind K,
680 const SCEV *Step, BinaryOperator *BOp)
681 : StartValue(Start), IK(K), Step(Step), InductionBinOp(BOp) {
682 assert(IK != IK_NoInduction && "Not an induction");
684 // Start value type should match the induction kind and the value
685 // itself should not be null.
686 assert(StartValue && "StartValue is null");
687 assert((IK != IK_PtrInduction || StartValue->getType()->isPointerTy()) &&
688 "StartValue is not a pointer for pointer induction");
689 assert((IK != IK_IntInduction || StartValue->getType()->isIntegerTy()) &&
690 "StartValue is not an integer for integer induction");
692 // Check the Step Value. It should be non-zero integer value.
693 assert((!getConstIntStepValue() || !getConstIntStepValue()->isZero()) &&
694 "Step value is zero");
696 assert((IK != IK_PtrInduction || getConstIntStepValue()) &&
697 "Step value should be constant for pointer induction");
698 assert((IK == IK_FpInduction || Step->getType()->isIntegerTy()) &&
699 "StepValue is not an integer");
701 assert((IK != IK_FpInduction || Step->getType()->isFloatingPointTy()) &&
702 "StepValue is not FP for FpInduction");
703 assert((IK != IK_FpInduction || (InductionBinOp &&
704 (InductionBinOp->getOpcode() == Instruction::FAdd ||
705 InductionBinOp->getOpcode() == Instruction::FSub))) &&
706 "Binary opcode should be specified for FP induction");
709 int InductionDescriptor::getConsecutiveDirection() const {
710 ConstantInt *ConstStep = getConstIntStepValue();
711 if (ConstStep && (ConstStep->isOne() || ConstStep->isMinusOne()))
712 return ConstStep->getSExtValue();
716 ConstantInt *InductionDescriptor::getConstIntStepValue() const {
717 if (isa<SCEVConstant>(Step))
718 return dyn_cast<ConstantInt>(cast<SCEVConstant>(Step)->getValue());
722 Value *InductionDescriptor::transform(IRBuilder<> &B, Value *Index,
724 const DataLayout& DL) const {
726 SCEVExpander Exp(*SE, DL, "induction");
727 assert(Index->getType() == Step->getType() &&
728 "Index type does not match StepValue type");
730 case IK_IntInduction: {
731 assert(Index->getType() == StartValue->getType() &&
732 "Index type does not match StartValue type");
734 // FIXME: Theoretically, we can call getAddExpr() of ScalarEvolution
735 // and calculate (Start + Index * Step) for all cases, without
736 // special handling for "isOne" and "isMinusOne".
737 // But in the real life the result code getting worse. We mix SCEV
738 // expressions and ADD/SUB operations and receive redundant
739 // intermediate values being calculated in different ways and
740 // Instcombine is unable to reduce them all.
742 if (getConstIntStepValue() &&
743 getConstIntStepValue()->isMinusOne())
744 return B.CreateSub(StartValue, Index);
745 if (getConstIntStepValue() &&
746 getConstIntStepValue()->isOne())
747 return B.CreateAdd(StartValue, Index);
748 const SCEV *S = SE->getAddExpr(SE->getSCEV(StartValue),
749 SE->getMulExpr(Step, SE->getSCEV(Index)));
750 return Exp.expandCodeFor(S, StartValue->getType(), &*B.GetInsertPoint());
752 case IK_PtrInduction: {
753 assert(isa<SCEVConstant>(Step) &&
754 "Expected constant step for pointer induction");
755 const SCEV *S = SE->getMulExpr(SE->getSCEV(Index), Step);
756 Index = Exp.expandCodeFor(S, Index->getType(), &*B.GetInsertPoint());
757 return B.CreateGEP(nullptr, StartValue, Index);
759 case IK_FpInduction: {
760 assert(Step->getType()->isFloatingPointTy() && "Expected FP Step value");
761 assert(InductionBinOp &&
762 (InductionBinOp->getOpcode() == Instruction::FAdd ||
763 InductionBinOp->getOpcode() == Instruction::FSub) &&
764 "Original bin op should be defined for FP induction");
766 Value *StepValue = cast<SCEVUnknown>(Step)->getValue();
768 // Floating point operations had to be 'fast' to enable the induction.
770 Flags.setUnsafeAlgebra();
772 Value *MulExp = B.CreateFMul(StepValue, Index);
773 if (isa<Instruction>(MulExp))
774 // We have to check, the MulExp may be a constant.
775 cast<Instruction>(MulExp)->setFastMathFlags(Flags);
777 Value *BOp = B.CreateBinOp(InductionBinOp->getOpcode() , StartValue,
778 MulExp, "induction");
779 if (isa<Instruction>(BOp))
780 cast<Instruction>(BOp)->setFastMathFlags(Flags);
787 llvm_unreachable("invalid enum");
790 bool InductionDescriptor::isFPInductionPHI(PHINode *Phi, const Loop *TheLoop,
792 InductionDescriptor &D) {
794 // Here we only handle FP induction variables.
795 assert(Phi->getType()->isFloatingPointTy() && "Unexpected Phi type");
797 if (TheLoop->getHeader() != Phi->getParent())
800 // The loop may have multiple entrances or multiple exits; we can analyze
801 // this phi if it has a unique entry value and a unique backedge value.
802 if (Phi->getNumIncomingValues() != 2)
804 Value *BEValue = nullptr, *StartValue = nullptr;
805 if (TheLoop->contains(Phi->getIncomingBlock(0))) {
806 BEValue = Phi->getIncomingValue(0);
807 StartValue = Phi->getIncomingValue(1);
809 assert(TheLoop->contains(Phi->getIncomingBlock(1)) &&
810 "Unexpected Phi node in the loop");
811 BEValue = Phi->getIncomingValue(1);
812 StartValue = Phi->getIncomingValue(0);
815 BinaryOperator *BOp = dyn_cast<BinaryOperator>(BEValue);
819 Value *Addend = nullptr;
820 if (BOp->getOpcode() == Instruction::FAdd) {
821 if (BOp->getOperand(0) == Phi)
822 Addend = BOp->getOperand(1);
823 else if (BOp->getOperand(1) == Phi)
824 Addend = BOp->getOperand(0);
825 } else if (BOp->getOpcode() == Instruction::FSub)
826 if (BOp->getOperand(0) == Phi)
827 Addend = BOp->getOperand(1);
832 // The addend should be loop invariant
833 if (auto *I = dyn_cast<Instruction>(Addend))
834 if (TheLoop->contains(I))
837 // FP Step has unknown SCEV
838 const SCEV *Step = SE->getUnknown(Addend);
839 D = InductionDescriptor(StartValue, IK_FpInduction, Step, BOp);
843 bool InductionDescriptor::isInductionPHI(PHINode *Phi, const Loop *TheLoop,
844 PredicatedScalarEvolution &PSE,
845 InductionDescriptor &D,
847 Type *PhiTy = Phi->getType();
849 // Handle integer and pointer inductions variables.
850 // Now we handle also FP induction but not trying to make a
851 // recurrent expression from the PHI node in-place.
853 if (!PhiTy->isIntegerTy() && !PhiTy->isPointerTy() &&
854 !PhiTy->isFloatTy() && !PhiTy->isDoubleTy() && !PhiTy->isHalfTy())
857 if (PhiTy->isFloatingPointTy())
858 return isFPInductionPHI(Phi, TheLoop, PSE.getSE(), D);
860 const SCEV *PhiScev = PSE.getSCEV(Phi);
861 const auto *AR = dyn_cast<SCEVAddRecExpr>(PhiScev);
863 // We need this expression to be an AddRecExpr.
865 AR = PSE.getAsAddRec(Phi);
868 DEBUG(dbgs() << "LV: PHI is not a poly recurrence.\n");
872 return isInductionPHI(Phi, TheLoop, PSE.getSE(), D, AR);
875 bool InductionDescriptor::isInductionPHI(PHINode *Phi, const Loop *TheLoop,
877 InductionDescriptor &D,
879 Type *PhiTy = Phi->getType();
880 // We only handle integer and pointer inductions variables.
881 if (!PhiTy->isIntegerTy() && !PhiTy->isPointerTy())
884 // Check that the PHI is consecutive.
885 const SCEV *PhiScev = Expr ? Expr : SE->getSCEV(Phi);
886 const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(PhiScev);
889 DEBUG(dbgs() << "LV: PHI is not a poly recurrence.\n");
893 if (AR->getLoop() != TheLoop) {
894 // FIXME: We should treat this as a uniform. Unfortunately, we
895 // don't currently know how to handled uniform PHIs.
896 DEBUG(dbgs() << "LV: PHI is a recurrence with respect to an outer loop.\n");
901 Phi->getIncomingValueForBlock(AR->getLoop()->getLoopPreheader());
902 const SCEV *Step = AR->getStepRecurrence(*SE);
903 // Calculate the pointer stride and check if it is consecutive.
904 // The stride may be a constant or a loop invariant integer value.
905 const SCEVConstant *ConstStep = dyn_cast<SCEVConstant>(Step);
906 if (!ConstStep && !SE->isLoopInvariant(Step, TheLoop))
909 if (PhiTy->isIntegerTy()) {
910 D = InductionDescriptor(StartValue, IK_IntInduction, Step);
914 assert(PhiTy->isPointerTy() && "The PHI must be a pointer");
915 // Pointer induction should be a constant.
919 ConstantInt *CV = ConstStep->getValue();
920 Type *PointerElementType = PhiTy->getPointerElementType();
921 // The pointer stride cannot be determined if the pointer element type is not
923 if (!PointerElementType->isSized())
926 const DataLayout &DL = Phi->getModule()->getDataLayout();
927 int64_t Size = static_cast<int64_t>(DL.getTypeAllocSize(PointerElementType));
931 int64_t CVSize = CV->getSExtValue();
934 auto *StepValue = SE->getConstant(CV->getType(), CVSize / Size,
936 D = InductionDescriptor(StartValue, IK_PtrInduction, StepValue);
940 bool llvm::formDedicatedExitBlocks(Loop *L, DominatorTree *DT, LoopInfo *LI,
941 bool PreserveLCSSA) {
942 bool Changed = false;
944 // We re-use a vector for the in-loop predecesosrs.
945 SmallVector<BasicBlock *, 4> InLoopPredecessors;
947 auto RewriteExit = [&](BasicBlock *BB) {
948 assert(InLoopPredecessors.empty() &&
949 "Must start with an empty predecessors list!");
950 auto Cleanup = make_scope_exit([&] { InLoopPredecessors.clear(); });
952 // See if there are any non-loop predecessors of this exit block and
953 // keep track of the in-loop predecessors.
954 bool IsDedicatedExit = true;
955 for (auto *PredBB : predecessors(BB))
956 if (L->contains(PredBB)) {
957 if (isa<IndirectBrInst>(PredBB->getTerminator()))
958 // We cannot rewrite exiting edges from an indirectbr.
961 InLoopPredecessors.push_back(PredBB);
963 IsDedicatedExit = false;
966 assert(!InLoopPredecessors.empty() && "Must have *some* loop predecessor!");
968 // Nothing to do if this is already a dedicated exit.
972 auto *NewExitBB = SplitBlockPredecessors(
973 BB, InLoopPredecessors, ".loopexit", DT, LI, PreserveLCSSA);
976 DEBUG(dbgs() << "WARNING: Can't create a dedicated exit block for loop: "
979 DEBUG(dbgs() << "LoopSimplify: Creating dedicated exit block "
980 << NewExitBB->getName() << "\n");
984 // Walk the exit blocks directly rather than building up a data structure for
985 // them, but only visit each one once.
986 SmallPtrSet<BasicBlock *, 4> Visited;
987 for (auto *BB : L->blocks())
988 for (auto *SuccBB : successors(BB)) {
989 // We're looking for exit blocks so skip in-loop successors.
990 if (L->contains(SuccBB))
993 // Visit each exit block exactly once.
994 if (!Visited.insert(SuccBB).second)
997 Changed |= RewriteExit(SuccBB);
1003 /// \brief Returns the instructions that use values defined in the loop.
1004 SmallVector<Instruction *, 8> llvm::findDefsUsedOutsideOfLoop(Loop *L) {
1005 SmallVector<Instruction *, 8> UsedOutside;
1007 for (auto *Block : L->getBlocks())
1008 // FIXME: I believe that this could use copy_if if the Inst reference could
1009 // be adapted into a pointer.
1010 for (auto &Inst : *Block) {
1011 auto Users = Inst.users();
1012 if (any_of(Users, [&](User *U) {
1013 auto *Use = cast<Instruction>(U);
1014 return !L->contains(Use->getParent());
1016 UsedOutside.push_back(&Inst);
1022 void llvm::getLoopAnalysisUsage(AnalysisUsage &AU) {
1023 // By definition, all loop passes need the LoopInfo analysis and the
1024 // Dominator tree it depends on. Because they all participate in the loop
1025 // pass manager, they must also preserve these.
1026 AU.addRequired<DominatorTreeWrapperPass>();
1027 AU.addPreserved<DominatorTreeWrapperPass>();
1028 AU.addRequired<LoopInfoWrapperPass>();
1029 AU.addPreserved<LoopInfoWrapperPass>();
1031 // We must also preserve LoopSimplify and LCSSA. We locally access their IDs
1032 // here because users shouldn't directly get them from this header.
1033 extern char &LoopSimplifyID;
1034 extern char &LCSSAID;
1035 AU.addRequiredID(LoopSimplifyID);
1036 AU.addPreservedID(LoopSimplifyID);
1037 AU.addRequiredID(LCSSAID);
1038 AU.addPreservedID(LCSSAID);
1039 // This is used in the LPPassManager to perform LCSSA verification on passes
1040 // which preserve lcssa form
1041 AU.addRequired<LCSSAVerificationPass>();
1042 AU.addPreserved<LCSSAVerificationPass>();
1044 // Loop passes are designed to run inside of a loop pass manager which means
1045 // that any function analyses they require must be required by the first loop
1046 // pass in the manager (so that it is computed before the loop pass manager
1047 // runs) and preserved by all loop pasess in the manager. To make this
1048 // reasonably robust, the set needed for most loop passes is maintained here.
1049 // If your loop pass requires an analysis not listed here, you will need to
1050 // carefully audit the loop pass manager nesting structure that results.
1051 AU.addRequired<AAResultsWrapperPass>();
1052 AU.addPreserved<AAResultsWrapperPass>();
1053 AU.addPreserved<BasicAAWrapperPass>();
1054 AU.addPreserved<GlobalsAAWrapperPass>();
1055 AU.addPreserved<SCEVAAWrapperPass>();
1056 AU.addRequired<ScalarEvolutionWrapperPass>();
1057 AU.addPreserved<ScalarEvolutionWrapperPass>();
1060 /// Manually defined generic "LoopPass" dependency initialization. This is used
1061 /// to initialize the exact set of passes from above in \c
1062 /// getLoopAnalysisUsage. It can be used within a loop pass's initialization
1065 /// INITIALIZE_PASS_DEPENDENCY(LoopPass)
1067 /// As-if "LoopPass" were a pass.
1068 void llvm::initializeLoopPassPass(PassRegistry &Registry) {
1069 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
1070 INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass)
1071 INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
1072 INITIALIZE_PASS_DEPENDENCY(LCSSAWrapperPass)
1073 INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)
1074 INITIALIZE_PASS_DEPENDENCY(BasicAAWrapperPass)
1075 INITIALIZE_PASS_DEPENDENCY(GlobalsAAWrapperPass)
1076 INITIALIZE_PASS_DEPENDENCY(SCEVAAWrapperPass)
1077 INITIALIZE_PASS_DEPENDENCY(ScalarEvolutionWrapperPass)
1080 /// \brief Find string metadata for loop
1082 /// If it has a value (e.g. {"llvm.distribute", 1} return the value as an
1083 /// operand or null otherwise. If the string metadata is not found return
1084 /// Optional's not-a-value.
1085 Optional<const MDOperand *> llvm::findStringMetadataForLoop(Loop *TheLoop,
1087 MDNode *LoopID = TheLoop->getLoopID();
1088 // Return none if LoopID is false.
1092 // First operand should refer to the loop id itself.
1093 assert(LoopID->getNumOperands() > 0 && "requires at least one operand");
1094 assert(LoopID->getOperand(0) == LoopID && "invalid loop id");
1096 // Iterate over LoopID operands and look for MDString Metadata
1097 for (unsigned i = 1, e = LoopID->getNumOperands(); i < e; ++i) {
1098 MDNode *MD = dyn_cast<MDNode>(LoopID->getOperand(i));
1101 MDString *S = dyn_cast<MDString>(MD->getOperand(0));
1104 // Return true if MDString holds expected MetaData.
1105 if (Name.equals(S->getString()))
1106 switch (MD->getNumOperands()) {
1110 return &MD->getOperand(1);
1112 llvm_unreachable("loop metadata has 0 or 1 operand");
1118 /// Returns true if the instruction in a loop is guaranteed to execute at least
1120 bool llvm::isGuaranteedToExecute(const Instruction &Inst,
1121 const DominatorTree *DT, const Loop *CurLoop,
1122 const LoopSafetyInfo *SafetyInfo) {
1123 // We have to check to make sure that the instruction dominates all
1124 // of the exit blocks. If it doesn't, then there is a path out of the loop
1125 // which does not execute this instruction, so we can't hoist it.
1127 // If the instruction is in the header block for the loop (which is very
1128 // common), it is always guaranteed to dominate the exit blocks. Since this
1129 // is a common case, and can save some work, check it now.
1130 if (Inst.getParent() == CurLoop->getHeader())
1131 // If there's a throw in the header block, we can't guarantee we'll reach
1133 return !SafetyInfo->HeaderMayThrow;
1135 // Somewhere in this loop there is an instruction which may throw and make us
1137 if (SafetyInfo->MayThrow)
1140 // Get the exit blocks for the current loop.
1141 SmallVector<BasicBlock *, 8> ExitBlocks;
1142 CurLoop->getExitBlocks(ExitBlocks);
1144 // Verify that the block dominates each of the exit blocks of the loop.
1145 for (BasicBlock *ExitBlock : ExitBlocks)
1146 if (!DT->dominates(Inst.getParent(), ExitBlock))
1149 // As a degenerate case, if the loop is statically infinite then we haven't
1150 // proven anything since there are no exit blocks.
1151 if (ExitBlocks.empty())
1154 // FIXME: In general, we have to prove that the loop isn't an infinite loop.
1155 // See http::llvm.org/PR24078 . (The "ExitBlocks.empty()" check above is
1156 // just a special case of this.)
1160 Optional<unsigned> llvm::getLoopEstimatedTripCount(Loop *L) {
1161 // Only support loops with a unique exiting block, and a latch.
1162 if (!L->getExitingBlock())
1165 // Get the branch weights for the the loop's backedge.
1166 BranchInst *LatchBR =
1167 dyn_cast<BranchInst>(L->getLoopLatch()->getTerminator());
1168 if (!LatchBR || LatchBR->getNumSuccessors() != 2)
1171 assert((LatchBR->getSuccessor(0) == L->getHeader() ||
1172 LatchBR->getSuccessor(1) == L->getHeader()) &&
1173 "At least one edge out of the latch must go to the header");
1175 // To estimate the number of times the loop body was executed, we want to
1176 // know the number of times the backedge was taken, vs. the number of times
1177 // we exited the loop.
1178 uint64_t TrueVal, FalseVal;
1179 if (!LatchBR->extractProfMetadata(TrueVal, FalseVal))
1182 if (!TrueVal || !FalseVal)
1185 // Divide the count of the backedge by the count of the edge exiting the loop,
1186 // rounding to nearest.
1187 if (LatchBR->getSuccessor(0) == L->getHeader())
1188 return (TrueVal + (FalseVal / 2)) / FalseVal;
1190 return (FalseVal + (TrueVal / 2)) / TrueVal;
1193 /// \brief Adds a 'fast' flag to floating point operations.
1194 static Value *addFastMathFlag(Value *V) {
1195 if (isa<FPMathOperator>(V)) {
1196 FastMathFlags Flags;
1197 Flags.setUnsafeAlgebra();
1198 cast<Instruction>(V)->setFastMathFlags(Flags);
1203 // Helper to generate a log2 shuffle reduction.
1205 llvm::getShuffleReduction(IRBuilder<> &Builder, Value *Src, unsigned Op,
1206 RecurrenceDescriptor::MinMaxRecurrenceKind MinMaxKind,
1207 ArrayRef<Value *> RedOps) {
1208 unsigned VF = Src->getType()->getVectorNumElements();
1209 // VF is a power of 2 so we can emit the reduction using log2(VF) shuffles
1210 // and vector ops, reducing the set of values being computed by half each
1212 assert(isPowerOf2_32(VF) &&
1213 "Reduction emission only supported for pow2 vectors!");
1214 Value *TmpVec = Src;
1215 SmallVector<Constant *, 32> ShuffleMask(VF, nullptr);
1216 for (unsigned i = VF; i != 1; i >>= 1) {
1217 // Move the upper half of the vector to the lower half.
1218 for (unsigned j = 0; j != i / 2; ++j)
1219 ShuffleMask[j] = Builder.getInt32(i / 2 + j);
1221 // Fill the rest of the mask with undef.
1222 std::fill(&ShuffleMask[i / 2], ShuffleMask.end(),
1223 UndefValue::get(Builder.getInt32Ty()));
1225 Value *Shuf = Builder.CreateShuffleVector(
1226 TmpVec, UndefValue::get(TmpVec->getType()),
1227 ConstantVector::get(ShuffleMask), "rdx.shuf");
1229 if (Op != Instruction::ICmp && Op != Instruction::FCmp) {
1230 // Floating point operations had to be 'fast' to enable the reduction.
1231 TmpVec = addFastMathFlag(Builder.CreateBinOp((Instruction::BinaryOps)Op,
1232 TmpVec, Shuf, "bin.rdx"));
1234 assert(MinMaxKind != RecurrenceDescriptor::MRK_Invalid &&
1236 TmpVec = RecurrenceDescriptor::createMinMaxOp(Builder, MinMaxKind, TmpVec,
1239 if (!RedOps.empty())
1240 propagateIRFlags(TmpVec, RedOps);
1242 // The result is in the first element of the vector.
1243 return Builder.CreateExtractElement(TmpVec, Builder.getInt32(0));
1246 /// Create a simple vector reduction specified by an opcode and some
1247 /// flags (if generating min/max reductions).
1248 Value *llvm::createSimpleTargetReduction(
1249 IRBuilder<> &Builder, const TargetTransformInfo *TTI, unsigned Opcode,
1250 Value *Src, TargetTransformInfo::ReductionFlags Flags,
1251 ArrayRef<Value *> RedOps) {
1252 assert(isa<VectorType>(Src->getType()) && "Type must be a vector");
1254 Value *ScalarUdf = UndefValue::get(Src->getType()->getVectorElementType());
1255 std::function<Value*()> BuildFunc;
1256 using RD = RecurrenceDescriptor;
1257 RD::MinMaxRecurrenceKind MinMaxKind = RD::MRK_Invalid;
1258 // TODO: Support creating ordered reductions.
1259 FastMathFlags FMFUnsafe;
1260 FMFUnsafe.setUnsafeAlgebra();
1263 case Instruction::Add:
1264 BuildFunc = [&]() { return Builder.CreateAddReduce(Src); };
1266 case Instruction::Mul:
1267 BuildFunc = [&]() { return Builder.CreateMulReduce(Src); };
1269 case Instruction::And:
1270 BuildFunc = [&]() { return Builder.CreateAndReduce(Src); };
1272 case Instruction::Or:
1273 BuildFunc = [&]() { return Builder.CreateOrReduce(Src); };
1275 case Instruction::Xor:
1276 BuildFunc = [&]() { return Builder.CreateXorReduce(Src); };
1278 case Instruction::FAdd:
1280 auto Rdx = Builder.CreateFAddReduce(ScalarUdf, Src);
1281 cast<CallInst>(Rdx)->setFastMathFlags(FMFUnsafe);
1285 case Instruction::FMul:
1287 auto Rdx = Builder.CreateFMulReduce(ScalarUdf, Src);
1288 cast<CallInst>(Rdx)->setFastMathFlags(FMFUnsafe);
1292 case Instruction::ICmp:
1293 if (Flags.IsMaxOp) {
1294 MinMaxKind = Flags.IsSigned ? RD::MRK_SIntMax : RD::MRK_UIntMax;
1296 return Builder.CreateIntMaxReduce(Src, Flags.IsSigned);
1299 MinMaxKind = Flags.IsSigned ? RD::MRK_SIntMin : RD::MRK_UIntMin;
1301 return Builder.CreateIntMinReduce(Src, Flags.IsSigned);
1305 case Instruction::FCmp:
1306 if (Flags.IsMaxOp) {
1307 MinMaxKind = RD::MRK_FloatMax;
1308 BuildFunc = [&]() { return Builder.CreateFPMaxReduce(Src, Flags.NoNaN); };
1310 MinMaxKind = RD::MRK_FloatMin;
1311 BuildFunc = [&]() { return Builder.CreateFPMinReduce(Src, Flags.NoNaN); };
1315 llvm_unreachable("Unhandled opcode");
1318 if (TTI->useReductionIntrinsic(Opcode, Src->getType(), Flags))
1320 return getShuffleReduction(Builder, Src, Opcode, MinMaxKind, RedOps);
1323 /// Create a vector reduction using a given recurrence descriptor.
1324 Value *llvm::createTargetReduction(IRBuilder<> &Builder,
1325 const TargetTransformInfo *TTI,
1326 RecurrenceDescriptor &Desc, Value *Src,
1328 // TODO: Support in-order reductions based on the recurrence descriptor.
1329 RecurrenceDescriptor::RecurrenceKind RecKind = Desc.getRecurrenceKind();
1330 TargetTransformInfo::ReductionFlags Flags;
1331 Flags.NoNaN = NoNaN;
1332 auto getSimpleRdx = [&](unsigned Opc) {
1333 return createSimpleTargetReduction(Builder, TTI, Opc, Src, Flags);
1336 case RecurrenceDescriptor::RK_FloatAdd:
1337 return getSimpleRdx(Instruction::FAdd);
1338 case RecurrenceDescriptor::RK_FloatMult:
1339 return getSimpleRdx(Instruction::FMul);
1340 case RecurrenceDescriptor::RK_IntegerAdd:
1341 return getSimpleRdx(Instruction::Add);
1342 case RecurrenceDescriptor::RK_IntegerMult:
1343 return getSimpleRdx(Instruction::Mul);
1344 case RecurrenceDescriptor::RK_IntegerAnd:
1345 return getSimpleRdx(Instruction::And);
1346 case RecurrenceDescriptor::RK_IntegerOr:
1347 return getSimpleRdx(Instruction::Or);
1348 case RecurrenceDescriptor::RK_IntegerXor:
1349 return getSimpleRdx(Instruction::Xor);
1350 case RecurrenceDescriptor::RK_IntegerMinMax: {
1351 switch (Desc.getMinMaxRecurrenceKind()) {
1352 case RecurrenceDescriptor::MRK_SIntMax:
1353 Flags.IsSigned = true;
1354 Flags.IsMaxOp = true;
1356 case RecurrenceDescriptor::MRK_UIntMax:
1357 Flags.IsMaxOp = true;
1359 case RecurrenceDescriptor::MRK_SIntMin:
1360 Flags.IsSigned = true;
1362 case RecurrenceDescriptor::MRK_UIntMin:
1365 llvm_unreachable("Unhandled MRK");
1367 return getSimpleRdx(Instruction::ICmp);
1369 case RecurrenceDescriptor::RK_FloatMinMax: {
1371 Desc.getMinMaxRecurrenceKind() == RecurrenceDescriptor::MRK_FloatMax;
1372 return getSimpleRdx(Instruction::FCmp);
1375 llvm_unreachable("Unhandled RecKind");
1379 void llvm::propagateIRFlags(Value *I, ArrayRef<Value *> VL, Value *OpValue) {
1380 auto *VecOp = dyn_cast<Instruction>(I);
1383 auto *Intersection = (OpValue == nullptr) ? dyn_cast<Instruction>(VL[0])
1384 : dyn_cast<Instruction>(OpValue);
1387 const unsigned Opcode = Intersection->getOpcode();
1388 VecOp->copyIRFlags(Intersection);
1389 for (auto *V : VL) {
1390 auto *Instr = dyn_cast<Instruction>(V);
1393 if (OpValue == nullptr || Opcode == Instr->getOpcode())
1394 VecOp->andIRFlags(V);